US 8233743 B2
The invention relates to a method and a device for detecting the orientation of an image in a set of images. All images in this set of images contain at least one similar object. The method proposes to choose one image in this set of images as being a reference image. The orientation of the other images is detected based on the orientation of this reference image.
1. Method for detecting the orientation of images in a set of captured images representing a similar scene, all the images in said set of captured images containing at least one similar object, wherein the method comprises the steps of:
choosing at least one reference image from the set of captured images, the reference image having an orientation that is known a priori,
wherein said choosing comprises selecting, for each target image in said set of captured images, a single reference image among a plurality of reference images whose orientations are known a priori, wherein said selected reference image is the reference image that has a minimum distance to said target image among the plurality reference images; and
detecting, by a hardware processor, orientation of said target image as a function of the orientation of said selected reference image.
2. Method according to
3. Method according to
4. Method according to
5. Method according to
6. Method according to
7. Method according to
8. Method according to
The invention relates to a device and a process for detecting the orientation of an image in a set of images.
This application claims the benefit, under 35 U.S.C. §365 of International Application PCT/EP05/050022, filed Jan. 4, 2005, which was published in accordance with PCT Article 21(2) on Jul. 28, 2005 in English and which claims the benefit of French patent applications No. 0400067, filed Jan. 6, 2004.
The invention relates more generally to the automatic detection of the orientation of an image that may possibly also contain text.
Like scanners, digital cameras generate photos which are often viewed on a screen, such as a computer screen. These images may be viewed correctly, that is to say in the right orientation or in a sense that does not allow the user to view them without having to turn his head to the left or to the right, or even to lower his head.
Specifically, sensing apparatus, such as digital cameras, can capture images in several senses, and in particular it is not uncommon for a user to take some of his photos in portrait mode and others in landscape mode. The photos thus taken are then transferred to a computer and all viewed in the same sense. Some will therefore be correctly viewed while others will require a 90, 180 or 270 degree rotation to be viewed correctly.
Certain sensing devices and in particular certain digital cameras possess orientation sensors that detect a rotation of the objective and transmit a rotation cue obtained from the sensor with the image. This allows the viewing device, by virtue of the rotation cue, to automatically perform a rotation of the image so that the latter appears in the correct sense.
Other devices use methods for extracting the low-level or high-level characteristics of the image. This makes it possible to analyse the content of the image in terms of colour, texture and also in terms of semantic content.
However, such devices are not robust for all types of images.
The invention relates more particularly to the automatic rotation of images in viewing devices not receiving any orientation cue from the sensing device and based on the orientation of other images.
Accordingly, the invention relates to a method for detecting the orientation of an image in a set of images, comprising the steps of:
It frequently happens that the user takes several images representing the same scene, certain images being taken in portrait mode and others in landscape mode. The detection of a reference image from among these images may make it possible to detect the orientation of the other images representative of the same scene or of a similar-content scene. This happens especially frequently in respect of images representing a person in a landscape, the user wishing in certain images to photograph the landscape and in others to make the person stand out by taking a close-up shot.
The invention will be better understood and illustrated by means of advantageous, wholly nonlimiting exemplary embodiments and modes of implementation, with reference to the appended drawings, in which:
As indicated in
The two subimages have the same relative size with respect to the image in which they are positioned. The two subimages have the same width/height ratio.
In the case where one or both subimages are smaller than the respective image in which they lie, the subimage is positioned in such a way as to obtain a minimum distance between the subimage of the reference image and the subimage of the target image whose orientation one wishes to detect. To this end, several positions of the subimages are tested and the best position is retained.
In another embodiment, the subimage is positioned at the centre of the image for the reference image and for the target image.
In another embodiment, the subimage is positioned according to a fixed rule depending on the class of the image. This presupposes that the image is classed previously according to a type which may for example be the type of scene (town, landscape, interior scene, person, etc.). It is possible to define a rule as follows: for landscape and interior scenes, the subimage is placed in the centre of the image; for town scenes, the subimage is placed at the bottom of the image.
In yet another embodiment, the subimage is placed according to a fixed rule depending on the presence and on the position of objects in the image. This involves the prior detection of objects in the image. Object can also include face. The rule possibly then being to centre the subimage on the object.
When the subimages are positioned in the reference image and in the target image, four measurements of visual distance are performed. The four measurements are performed:
To do this, the subimages of the reference image and of the target image are sliced up into blocks A*B, A vertically and B horizontally.
This slicing is aimed at complying with the non-stationarity of the images processed (the various zones of the images exhibit very different contents). The higher the number of blocks, the more this effect will be taken into account.
The number A*B should not however exceed a certain threshold depending on the resolution of the images in order to contain enough information.
The four measurements of distance D between the target image and reference image are calculated thus:
The distances Di represent the four distances D0, D1, D2, D3 between the four blocks of the images, as indicated in
A weighting factor wi is applied for each block. The weighting can depend either on a classification of the blocks according to their significance or on a measure of confidence in respect of the calculation of the distances.
Each distance Di is calculated from M particular distances dm according to the following formula:
m representing the number of descriptors used for the calculation of a distance. The descriptors possibly being colour, texture for example.
The calculation of the distance dm is carried out using processes known to the person skilled in the art. It is possible to effect colour histogram differences, contour histogram differences, the difference of subband energies, the difference of responses of directional filters so as to measure these distances.
A weighting factor vm is applied for each distance dm so as to take account of the confidence of calculation of the m distances.
The reference images are images whose orientation is known a priori. The orientation of these images may be obtained through various methods, and in particular by detecting for example objects in these images or faces, but it is also possible to index them manually by the user. It is also possible to use methods different from the method proposed by the present invention making it possible to detect the orientation of an image and these methods may be used to detect the orientation of the reference image. Such a method is described for example in European Patent Application 1107182 filed in the name of the company Eastman Kodak Company on 17 Nov. 2000 and entitled “determining orientation of images containing blue sky”.
When several reference images are available to the method, the choice of the reference image is performed by making a calculation of J distances Ej between the given image and the reference images available to the method.
Each distance Ej is calculated from L particular distances el according to the following formula:
A weighting factor ql is applied to each distance el calculated.
Each distance el calculated is calculated according to known processes, as indicated previously in respect of the distances dm.
The reference image for which the distance with the target image is the smallest is then chosen.
Subsequently the measurements of distances Di such as indicated previously are calculated as is D.
D is therefore calculated between the target image and the reference image in four positions (turned by 0, 90, 180 or 270 degrees).
The smallest distance indicates the rotation that the target image must undergo in order to be viewed in the correct sense. For example, if the smallest distance is that between the target image and the reference image turned by 270 degrees, then the target image must be turned by 207 degrees so as to be viewed in the correct sense.